Submission 83
ValiDATA AI
Submitter group: Companies.
An AI model identified the submitter group and wrote the summary, key points and recommendation summaries. Counts do not measure importance or merit.
Summary
Read the original submission for the full text.
ValiDATA AI, an AI governance consultancy whose agentic commerce work is conceptual, examines agents that place orders and initiate payments. Recommends coordinated legal review, consistent regulator guidance, AI Safety Institute testing, guidance on transaction records and privacy, and domestic assurance capability, building on existing law.
Key points
Quotes are the submitter’s own words. Text taken from documents, including scanned pages read by text-recognition software, can differ from the PDF. Check the original. Listed in the same order as on the submission’s page in the interactive guide.
ValiDATA states that the practical questions for agentic commerce are who authorised an action, what limits applied, what evidence exists and how an error can be remedied.
For consumers and businesses, the practical questions are who authorised an action, what limits applied, what evidence exists and how an error can be remedied.
ValiDATA states that a review should identify which party must establish authority, what evidence is needed and which remedy is available, as questions for examination and not findings that existing law lacks a remedy.
The review should identify which party must establish authority, what evidence is needed, and which remedy is available. These are questions for examination, not findings that existing law provides no remedy.
ValiDATA proposes governance questions as control-design considerations proportionate to transaction consequences, and says the synthesis has not been operationally validated by it in commercial-agent deployments.
The questions below are proposed control-design considerations, proportionate to the consequences of the transaction. The synthesis is intended to guide control design and assessment; it has not been operationally validated by ValiDATA in commercial-agent deployments.
ValiDATA recommends connecting the existing initiatives it describes through a defined set of commercial scenarios and published explanations of regulatory responsibilities.
ValiDATA AI recommends connecting these initiatives through a defined set of commercial scenarios and published explanations of regulatory responsibilities.
ValiDATA says agentic commerce raises questions about the location, access and availability of transaction records, which may include sensitive and personal information.
For term (f), agentic commerce raises questions about the records needed to establish what was authorised and executed. Those records may include commercially sensitive information and personal information. Their location, access arrangements and availability during a dispute should form part of the governance assessment.
ValiDATA proposes Australian capability in agent evaluation and assurance as a priority, through research partnerships, testing methods and skills, without requiring all models or data be hosted domestically.
For term (e), we recommend supporting Australian capability in agent evaluation, secure systems integration and independent assurance. Practical research partnerships, reusable testing methods and skills development would help Australian businesses evaluate commercial agents and retain the ability to investigate their operation. This is a proposed capability priority, not a claim that all models or transaction data must be hosted domestically.
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
Distinguish decision assistance from authority to commit to transactions, and distinguish consumer purchasing, business procurement and regulated financial services in guidance.
Guidance should distinguish assistance with a decision from authority to commit someone to a transaction. It should also distinguish consumer purchasing, business procurement and regulated financial services, where protections and approval processes may differ.
Have the consumer-law workstream publish worked scenarios on delegated authority, contract formation, errors, liability and remedies, including financial-services and state and territory law, and identify demonstrated gaps before proposing new obligations.
Recommend that the Government's announced consumer-law workstream publish worked scenarios covering delegated authority, contract formation, transaction errors, liability and remedies. Include relevant financial-services and state and territory law, and identify demonstrated gaps before proposing new obligations.
Test protections against exceeded instructions, apparent authorisation, agent-generated errors without human input errors, and disputed payments after supply; identify burdens of establishing authority, required evidence and available remedies.
We recommend testing these existing protections against concrete scenarios: an agent exceeding instructions; a merchant relying on an apparent authorisation; an incorrect order generated without a human input error; and a payment disputed after goods have been supplied. The review should identify which party must establish authority, what evidence is needed, and which remedy is available. These are questions for examination, not findings that existing law provides no remedy.
Explain how technical mechanisms relate to legal authority and remedies; treat successful technical checks as only part of the evidence of permission and dispute handling.
In our view, public guidance should explain how such mechanisms relate to legal authority and remedies. A successful technical check is one part of the evidence to consider; organisations still need to establish what the agent was permitted to do and how a dispute will be handled.
Have Treasury coordinate the relevant regulators to address the five questions in section 3, explain agency responsibilities and distinguish legal duties from recommended practices, with financial-system issues drawing on Council of Financial Regulators arrangements.
Recommend that Treasury coordinate the relevant regulators to address the five questions in section 3, explain agency responsibilities and distinguish legal duties from recommended practices. Financial-system issues should draw on existing Council of Financial Regulators arrangements.
Address agent identity, representation, operational responsibility, distinguishing organisational/supplier/customer agents and revoking compromised credentials in practical governance guidance.
Who is acting? Identify the agent, the person or organisation it represents, and the person responsible for its operation. Organisations should be able to distinguish their own agents from those of suppliers and customers, and revoke compromised credentials.
Address delegated negotiation, ordering, terms and payment authority, including grantor, duration, change and revocation; distinguish identity checks from permission.
What is the agent authorised to do? Specify the activities delegated, including any permission to negotiate, order, accept terms or initiate payment. Record who granted that permission, its duration and how it can be changed or withdrawn. Passing an identity check should not be treated as permission for every action.
Address counterparties, spending/frequency limits and human approvals; enforce restrictions through transaction software and test ambiguous instructions, misleading information and compromised services.
What limits apply? Define permitted counterparties, spending and frequency limits, and decisions requiring human approval. Enforce relevant restrictions through the surrounding software and transaction systems. Test how those controls behave when the agent encounters ambiguous instructions, misleading information or a compromised service.
Address proportionate, protected, understandable transaction records linking instructions, permission, terms, approvals and outcomes, including timestamps/system versions; do not treat agent explanations alone as execution proof.
What record demonstrates what happened? Preserve proportionate records linking the instruction, delegated permission, transaction terms, approval events and execution outcome. Include timestamps and relevant system versions. Records should be protected against alteration and understandable to those investigating a dispute; an agent's explanation alone should not be treated as proof of execution.
Address responsible-person investigation, stopping further action and remedies; explain consumer disputes, support cancellation/reversal where available and preserve dispute evidence.
What happens when something goes wrong? Provide a clear route to a responsible person who can investigate, stop further action and arrange an appropriate remedy. Consumers should receive understandable information about how to raise a dispute. Systems should support cancellation or reversal where available and preserve evidence when a dispute arises.
Connect commercial scenarios and published regulatory responsibilities through Treasury, ACCC, ASIC, APRA, OAIC and state/territory authorities, drawing on the Council of Financial Regulators where relevant and AISI/ASD technical work.
ValiDATA AI recommends connecting these initiatives through a defined set of commercial scenarios and published explanations of regulatory responsibilities. Treasury's consumer-law workstream could coordinate with the ACCC, ASIC, APRA, the OAIC and relevant state and territory authorities, drawing on the Council of Financial Regulators where financial-system issues arise. Technical input should build on the AI Safety Institute and ASD's work.
Ask the AI Safety Institute, with ASD and regulators, to test purchasing, payment and agent-to-agent interactions, and publish methods and findings where appropriate, including the limits of tested safeguards.
Recommend that the Government ask the AI Safety Institute, working with ASD and relevant regulators, to build on existing research through tests of purchasing, payment and interactions between agents. Publish methods and findings where appropriate, including the limits of tested safeguards.
Test compromised credentials, malicious instructions and inter-agent failures affecting payments or essential supply chains; distinguish demonstrated incidents, controlled tests and hypothetical risks.
For terms (k), (l) and (m), testing should examine how compromised credentials, malicious instructions and failures between agents could affect payments or essential supply chains. Findings should distinguish demonstrated incidents, controlled tests and hypothetical risks.
Have relevant regulators, including the OAIC, develop integrated transaction-evidence and overseas-arrangement guidance, with automated-decision transparency worked examples where applicable.
Recommend that the relevant regulators, including the OAIC, develop guidance on evidence integrity, access, proportionate retention, overseas arrangements and provider exit. Include worked examples of the automated-decision transparency provisions where applicable.
Guide proportionate retention, tamper protection, retrieval, access rights and provider-change/closure transfers; explain jurisdiction/provider access barriers and risk assessment, respecting privacy and lawful retention limits.
We recommend guidance on proportionate retention, tamper protection, retrieval, access rights and transfer of records when providers change or cease operating. It should explain when jurisdiction or provider arrangements may impede access, and how to assess that risk. Recordkeeping should satisfy evidentiary needs while respecting privacy and lawful retention limits.
Provide worked examples of when transacting agents meet automated-decision transparency significance/connection tests, distinguishing disclosure from transaction-specific consent and dispute resolution.
The privacy-policy transparency obligations commencing on 10 December 2026 are also relevant where covered entities arrange automated decisions using personal information that meet the statutory significance and connection tests.15 We recommend worked examples explaining when agents that transact fall within these provisions. This disclosure duty should be distinguished from transaction-specific consent and dispute resolution.
Support Australian agent evaluation, secure systems integration and independent assurance through research partnerships, reusable tests and skills; this does not require all models or transaction data to be domestically hosted.
For term (e), we recommend supporting Australian capability in agent evaluation, secure systems integration and independent assurance. Practical research partnerships, reusable testing methods and skills development would help Australian businesses evaluate commercial agents and retain the ability to investigate their operation. This is a proposed capability priority, not a claim that all models or transaction data must be hosted domestically.
Have the National AI Centre work with regulators, researchers and industry on reusable procurement questions, documented-delegation examples, testing methods and accessible guidance for smaller businesses, prioritising Australian evaluation and assurance skills, with controls scaled to transaction consequences.
Recommend that the National AI Centre work with regulators, researchers and industry on reusable procurement questions, examples of documented delegations, testing methods and accessible guidance for smaller businesses. Prioritise Australian evaluation and assurance skills, with controls scaled to the transaction's consequences.
Have the Office of AI work with relevant regulators on interim guidance distinguishing existing obligations, contractual protections and remaining infrastructure-dependency risks, and explaining their interaction with the proposed Australian Standards for AI.
Recommend that the Office of AI work with relevant regulators on interim guidance distinguishing existing obligations, contractual protections and remaining infrastructure-dependency risks, and explaining their interaction with the proposed Australian Standards for AI.